Simulation of Layout-Dependent STI Stress and Its Impact on Circuit Performance

Abstract

The impact of STI stress with layout dependency on circuit performance is investigated. A 3D stress simulator has been developed using finite element method, which considers both the layout design and process information (PDK). The mobility change due to stress is included in the transistor modeling for circuit simulation. The circuit performance can thus be analyzed with nonlocal stress. As a test case, a buffered SR flip-flop was simulated with and without STI stress considered. It can be seen that STI stress has non-negligible influence on the circuit performance.

Full text

Page 1

Simulation of Layout-Dependent STI Stress and Its
Impact on Circuit Performance
Liu Yang, Xiaojian Li, Lilin Tian and Zhiping Yu
Institute of Microelectronics
Tsinghua University
Beijing, China 100084
Email: yangliu@mails.thu.edu.cn
Abstract—The impact of STI stress with layout dependency
on circuit performance is investigated. A 3D stress simulator has
been developed using finite element method, which considers both
the layout design and process information (PDK). The mobility
change due to stress is included in the transistor modeling for
circuit simulation. The circuit performance can thus be analyzed
with nonlocal stress. As a test case, a buffered SR flip-flop was
simulated with and without STI stress considered. It can be
seen that STI stress has non-negligible influence on the circuit
performance.
I. INTRODUCTION
Various stress engineering techniques have been introduced
to improve the device performance as the scaling down of
CMOS devices continues. However, stresses can also gen-
erate in the device channels during the fabrication process.
Shallow trench isolation (STI) is one of the major sources
of such stress-inducing mechanisms, the magnitude of STI
stress depends on the layout pattern [1]. The performance of
a transistor is therefor related to its surrounding layout, and is
the performance of an entire circuit.
Efforts have been made to model the layout-dependent
stress and to exploit the possibility of managing it as a
performance enhancing technique [2]–[5]. Almost all these
works were based on 1D layout parameters or pseudo-2D,
“ideal” MOSFETs with rectangular or piecewise-rectangular
active regions were assumed in these studies. The effects of
neighboring devices were also not fully incorporated. How-
ever, such simplified model is different from the practical
situations. As shown in Fig.4, even the layout of a simple
standard cell contains complex neighboring polygon regions.
As the layout is essentially two-dimensional, detailed analysis
of STI stress should take into account 2D in-plane effects
rather than 1D effects only. Therefore, taking into account
the thickness direction, three-dimensional simulations must be
conducted to show the stress distribution correctly in numerical
analysis. Elaborated TCAD simulators have been developed
to perform the simulations for stress introduced by different
technology processes. Unfortunately, as limited by the amount
of memory and computational power available, currently such
TCAD tools can only be applied directly in the simulations
for layouts containing no more than a couple of transistors.
To perform numerical simulations for the stress profiles of
larger cells, certain simplification together with appropriate
methodology must be introduced to accelerate the simulation
process. New simulation tools should be developed to enable
the simulation of larger scale circuits.
In this work, a 3D parallel finite element simulator capable
of the simulations of larger circuits containing dozens of
transistors was developed to evaluate the layout-dependent
stress. Comparisons between simulated result obtained from
our program and experimental measurement were made. The
effects of layout-dependent stress on the performance of
an SR flip-flop standard cell are analyzed as an example.
Circuit simulations were performed with or without the layout-
dependent stress effect for circuit parameters to investigate the
impact of STI stress.
II. SIMULATION APPROACH
For the STI process, the major stress sources and effects
include thermal mismatch, intrinsic stress, growth of materials
and viscoelasticity. The thermal mismatch stress is resulted
from the difference of thermal expansion properties between
different materials such as silicon and silicon dioxide, it
can be calculated from the temperature change and thermal
expansion coefficients; Thin films such as nitride film are
generally highly stressed and thus have important effect on
the overall stress distribution, an initial stress is included
to model this effect; Volume expands during the oxidation
process for oxide liner in the STI process, thus additional
mechanical stress accumulates. This effect also contributes its
part to the initial strain of the liner layer; Some materials,
for example, nitride and silicon dioxide as concerned in our
study, exhibits noticeable viscoelasticity at higher temperature
and can not be neglected, a relaxation process is applied in
reference to temperature change for simplification. Beside all
these stress-inducing mechanisms, an additional initial stress is
also introduced as calibration to model a practical fabrication
technology.
The stress profile is solved using finite element method
(FEM). As the direct solution of the whole system is inappli-
cable due to the large size of mesh, the problem is partitioned
into smaller sub-regions and solved iteratively, which naturally
enables parallel computing.
Piezoresistance model was employed to calculate the in-
fluence of stress on mobility since it is effective for small
stress and can give the enhancement factor directly [6]. Model
978-1-4244-3947-8/09/$25.00 ©2009 IEEE 281

Page 2

Fig. 1. Schematical structure of the test sample [8]. Width of the nitride/poly-
Si is 9.4μm.
parameters presented in [7] were used. The mobility value
for every transistor is then calculated from stress tensors and
passed to a circuit simulator, then the circuit performance can
be obtained from circuit simulations.
III. RESULTS AND DISCUSSIONS
A simulation program was developed to implement the ap-
proach introduced previously. The simulation program consists
of two components: the layout analyzer and the finite element
simulator. In the layout analyzer, a modified scanning line
algorithm was employed to process the layout geometries and
extract the gate region for each transistor, then the whole
layout was partitioned into smaller “effective regions” for FEM
analysis. In the finite element simulator, first order Manhattan
bricks were used for 3D meshing. The distribution and size of
the elements were controlled along different directions sepa-
rately to generate high quality meshes. In the solving phase,
the resulting linear equations were solved using Generalized
Minimal Residual Method (GMRES).
As a validation of our simulation approach, A test structure
was simulated first to give a comparison between the simu-
lation result of our program and experimental measurement.
After the comparison, a buffered SR flip-flop standard cell
was simulated to evaluate the impact of STI stress on the
performance of practical circuits.
A. Comparisons
Micro-Raman spectroscopy was applied to determine me-
chanical stress in devices and structures used in microelectron-
ics in [8]. One of the samples measured was shown in Fig.1.
A 10-nm-thick pad oxide is grown on (001) silicon wafers by
dry oxidation at 900◦C. This is followed by deposition of 50
nm poly-amorphous silicon with intrinsic stress of -0.3GPa.
Next a low-pressure chemical vapor deposition (LPCVD)
silicon nitride film with intrinsic stress of 1.2GPa is deposited
and photolithographically patterned into long [-110]-oriented
nitride/poly-Si lines. The thickness of the nitride film is 240
nm. The width of the nitride/poly-Si is 9.4μm.
The same structure was simulated using our program and the
stress profile was obtained. Fig.2 shows the stress component
σxx in silicon near and beneath the upper silicon surface
obtained from the simulation, together with the experimental
measurements reported in [8]. Although slight differences are
shown near the boundaries of the nitride/poly-Si line, good
overall agreement is shown between the simulation result and
the measurement result.
Fig. 2. Comparison between experimental measurements result from [8] and
simulation results of σxx using our program for the same structure shown in
Fig.1.
Fig. 3. Schematic of the buffered SR flip-flop.
B. Simulation Results
To apply our simulation approach to practical circuit, a
0.18μm buffered SR flip-flop standard cell with minimum
size was simulated. The schematic of the circuit is shown
in Fig.3. Its layout is given in Fig.4 with only some layers
shown. The stress simulation was performed first to obtain
the layout-dependent stress profile for each transistor. The
corresponding degrees of freedom for the entire layout is
4,643,856. Fig.5 gives the in-plane result of stress tensor
component σxx near the upper surface of silicon, across the
Fig. 4.
and one metal layer included.
Layout of the buffered SR flip-flop with active layer, poly-Si layer
978-1-4244-3947-8/09/$25.00 ©2009 IEEE 282

Page 3

Fig. 5.
silicon surface.
In-plane Stress profile of σxx corresponding to the layout near the
Fig. 6.
the nitride and oxide layer removed and meshes shown.
3-D stress profile of σxx for a sub-region of the whole circuit, with
whole layout region. The magnitudes of σxxare depicted by
different colors. The effects of different active region shapes
on stress distribution can be seen in the visualization of stress
distribution. It is shown obviously that narrower active regions
result higher stress and wider active regions introduce smaller
stress changes. Meanwhile, the stress profiles near corners of
active region exhibit clear nonuniform 2D distributions, such
distributions can not be predicted in 2D simulations.
Fig.6 illustrates the 3-D result of one sub region, surface
colors correspond to the magnitudes of σxx as well. The
meshes are deformed exaggeratedly in reference of the nodal
displacements to give a better view. The transition of color
on the top surface indicates obvious 2D in-plane variation of
stress.
As the stress value differs along the width direction for each
MOSFET, but only one scalar mobility value can be accepted
by the device model, an integration operation was performed
along the central line in width direction for each MOSFET to
give the average stress values. This value was then passed to
the piezoresistance mobility model and the mobility value was
calculated for each transistor.
Fig. 7.
effect.
Timing parameters of the SR flip-flop with or without STI stress
With these new mobility values containing the influence of
layout-dependent stress, circuit simulations can be performed
to show the changes of circuit properties. The simulation
tool HSPICE was used for these simulations. Performance
parameters, including rise time tr, fall time tfand propagation
delays tpHL, tpLHwere obtained from the simulations. While
the rise time tr increased from 1.899ns without the layout-
dependent stress effect to 2.069ns with the layout-dependent
stress effect, fall time tf increased from 0.483ns to 0.587ns.
Parameters tpHLand tpLHare shown in Fig.7, with or without
the STI stress. A 8.9% increase of tpLHand a 14.6% increase
of tpHL were observed in circuit simulations due to the
inclusion of STI stress, as shown in Fig.7. These simulation
results indicate that the layout dependent stress has noticeable
impacts on the device properties and thus the performance of
the whole circuits.
IV. CONCLUSION
A numerical simulation program has been developed for 3D
stress analysis. This program was applied in the analysis of
practical devices. Comparisons between the simulation result
and experimental measurement of stress distribution validated
the accuracy of the simulation program. Simulation for the
impact of layout-dependent stress on circuit performance was
conducted using an SR flip-flop design. It is concluded that the
global stress has non-negligible effects on circuit performance.
Our program provides a good simulation tool for quantitative
evaluation of the layout-dependent STI stress and its impact
on circuit performance.
REFERENCES
[1] R. A. Bianchi, G. Bouche, and O. Roux-dit-Buisson, “Accurate model-
ing of trench isolation induced mechanical stress effects on MOSFET
electrical performance,” in IEDM, 2002, pp. 117–120.
[2] V. Moroz and et al., “The impact of layout on stress-enhanced transistor
performance,” in SISPAD, 2005, pp. 143–146.
[3] A. B. Kahng, P. Sharma, and R. O. Topaloglu, “Exploiting STI stress for
performance,” in ICCAD, 2007, pp. 83–90.
[4] P. W. Liu and et al., “Improved layout dependence in high performance
SiGe channel CMOSFETs,” in VLSI-TSA, 2008, pp. 161–162.
[5] A. B. Kahng and et al., “Chip optimization through STI-stress-aware
placement perturbations and fill insertion,” IEEE Trans. ICSCAD, vol. 27,
no. 7, pp. 1241–1252, 2008.
978-1-4244-3947-8/09/$25.00 ©2009 IEEE 283

Page 4

[6] C. S. Smith, “Piezoresistance effect in germanium and silicon,” Physical
review, vol. 94, no. 1, pp. 42–49, 1954.
[7] C. Gallon and et al., “Electrical analysis of mechanical stress induced
by STI in short MOSFETs using externally applied stress,” IEEE Trans.
Electron Devices, vol. 51, no. 8, pp. 1254–1261, 2004.
[8] I. D. Wolf, H. E. Maes, and S. K. Jones, “Stress measurements in silicon
devices through raman spectroscopy: Bridging the gap between theory
and experiment,” J. Appl. Phys., vol. 79, no. 9, pp. 7148–7156, 1996.
978-1-4244-3947-8/09/$25.00 ©2009 IEEE284